## datatable function from DT package create an HTML widget display of the dataset
## install DT package if the package is not yet available in your R environment
readxl::read_excel("dataset/dataset-variable-description.xlsx") |>
DT::datatable()Insert title here
BCon 147: special topics
1 Project overiew
In this project, we will explore employee attrition and performance using the HR Analytics Employee Attrition & Performance dataset. The primary goal is to develop insights into the factors that contribute to employee attrition. By analyzing a range of factors, including demographic data, job satisfaction, work-life balance, and job role, we aim to help businesses identify key areas where they can improve employee retention.
2 Scenario
Imagine you are working as a data analyst for a mid-sized company that is experiencing high employee turnover, especially among high-performing employees. The company has been facing increased costs related to hiring and training new employees, and management is concerned about the negative impact on productivity and morale. The human resources (HR) team has collected historical employee data and now looks to you for actionable insights. They want to understand why employees are leaving and how to retain talent effectively.
Your task is to analyze the dataset and provide insights that will help HR prioritize retention strategies. These strategies could include interventions like revising compensation policies, improving job satisfaction, or focusing on work-life balance initiatives. The success of your analysis could lead to significant cost savings for the company and an increase in employee engagement and performance.
3 Understanding data source
The dataset used for this project provides information about employee demographics, performance metrics, and various satisfaction ratings. The dataset is particularly useful for exploring how factors such as job satisfaction, work-life balance, and training opportunities influence employee performance and attrition.
This dataset is well-suited for conducting in-depth analysis of employee performance and retention, enabling us to build predictive models that identify the key drivers of employee attrition. Additionally, we can assess the impact of various organizational factors, such as training and work-life balance, on both performance and retention outcomes.
4 Data wrangling and management
Libraries
Before we start working on the dataset, we need to load the necessary libraries that will be used for data wrangling, analysis and visualization. Make sure to load the following libraries here. For packages to be installed, you can use the install.packages function. There are packages to be installed later on this project, so make sure to install them as needed and load them here.
# load all your libraries here
library(tidyverse)
library(readxl)
library(janitor)
library(tidytext)4.1 Data importation
Import the two dataset
Employee.csvandPerformanceRating.csv. Save theEmployee.csvasemployee_dtaandPerformanceRating.csvasperf_rating_dta.Merge the two dataset using the
left_joinfunction fromdplyr. Use theEmployeeIDvariable as the varible to join by. You may read more information about theleft_joinfunction here.Save the merged dataset as
hr_perf_dtaand display the dataset using thedatatablefunction fromDTpackage.
## import the two data here
employee_dta <- read_csv("dataset/Employee.csv")
perf_rating_dta <- read_csv("dataset/PerformanceRating.csv")
## merge employee_dta and perf_rating_dta using left_join function.
## save the merged dataset as hr_perf_dta
hr_perf_dta <-
employee_dta |>
left_join(perf_rating_dta, by = "EmployeeID") |>
mutate(bi_attrition = if_else(Attrition == "No", 0, 1))
## Use the datatable from DT package to display the merged dataset
DT::datatable(hr_perf_dta)4.2 Data management
Using the
clean_namesfunction fromjanitorpackage, standardize the variable names by using the recommended naming of variables.Save the renamed variables as
hr_perf_dtato update the dataset.
## clean names using the janitor packages and save as hr_perf_dta
hr_perf_dta <-
hr_perf_dta |>
clean_names()
## display the renamed hr_perf_dta using datatable function
DT::datatable(hr_perf_dta)Create a new variable
cat_educationwhereineducationis1=No formal education;2=High school;3=Bachelor;4=Masters;5=Doctorate. Use thecase_whenfunction to accomplish this task.Similarly, create new variables
cat_envi_sat,cat_job_sat, andcat_relation_satforenvironment_satisfaction,job_satisfaction, andrelationship_satisfaction, respectively. Re-code the values accordingly as1=Very dissatisfied;2=Dissatisfied;3=Neutral;4=Satisfied; and5=Very satisfied.Create new variables
cat_work_life_balance,cat_self_rating,cat_manager_ratingforwork_life_balance,self_rating, andmanager_rating, respectively. Re-code accordingly as1=Unacceptable;2=Needs improvement;3=Meets expectation;4=Exceeds expectation; and5=Above and beyond.Create a new variable
bi_attritionby transformingattritionvariable as a numeric variabe. Re-code accordingly asNo=0, andYes=1.Save all the changes in the
hr_perf_dta. Note that saving the changes with the same name will update the dataset with the new variables created.
## create cat_education
hr_perf_dta <-
hr_perf_dta |>
mutate(cat_education = case_when(
education == 1 ~ "No formal education",
education == 2 ~ "High school",
education == 3 ~ "Bachelor",
education == 4 ~ "Masters",
education == 5 ~ "Doctorate"
))
## create cat_envi_sat, cat_job_sat, and cat_relation_sat
hr_perf_dta <-
hr_perf_dta |>
mutate(cat_envi_sat = case_when(
environment_satisfaction == 1 ~ "Very dissatisfied",
environment_satisfaction == 2 ~ "Dissatisfied",
environment_satisfaction == 3 ~ "Neutral",
environment_satisfaction == 4 ~ "Satisfied",
environment_satisfaction == 5 ~ "Very satisfied"
)) |>
mutate(cat_job_sat = case_when(
job_satisfaction == 1 ~ "Very dissatisfied",
job_satisfaction == 2 ~ "Dissatisfied",
job_satisfaction == 3 ~ "Neutral",
job_satisfaction == 4 ~ "Satisfied",
job_satisfaction == 5 ~ "Very satisfied"
)) |>
mutate(cat_relation_sat = case_when(
relationship_satisfaction == 1 ~ "Very dissatisfied",
relationship_satisfaction == 2 ~ "Dissatisfied",
relationship_satisfaction == 3 ~ "Neutral",
relationship_satisfaction == 4 ~ "Satisfied",
relationship_satisfaction == 5 ~ "Very satisfied"
))
## create cat_work_life_balance, cat_self_rating, and cat_manager_rating
hr_perf_dta <-
hr_perf_dta |>
mutate(cat_work_life_balance = case_when(
work_life_balance == 1 ~ "Unacceptable",
work_life_balance == 2 ~ "Needs improvement",
work_life_balance == 3 ~ "Meets expectation",
work_life_balance == 4 ~ "Exceeds expectation",
work_life_balance == 5 ~ "Above and beyond"
)) |>
mutate(cat_self_rating = case_when(
self_rating == 1 ~ "Unacceptable",
self_rating == 2 ~ "Needs improvement",
self_rating == 3 ~ "Meets expectation",
self_rating == 4 ~ "Exceeds expectation",
self_rating == 5 ~ "Above and beyond"
)) |>
mutate(cat_manager_rating = case_when(
manager_rating == 1 ~ "Unacceptable",
manager_rating == 2 ~ "Needs improvement",
manager_rating == 3 ~ "Meets expectation",
manager_rating == 4 ~ "Exceeds expectation",
manager_rating == 5 ~ "Above and beyond"
))
## create bi_attrition
hr_perf_dta <-
hr_perf_dta |>
mutate(bi_attrition = if_else(attrition == "No", 0, 1))
## print the updated hr_perf_dta using datatable function
DT::datatable(hr_perf_dta)5 Exploratory data analysis
5.1 Descriptive statistics of employee attrition
Select the variables
attrition,job_role,department,age,salary,job_satisfaction, andwork_life_balance.Save asattrition_key_var_dta.Compute and plot the attrition rate across
job_role,department, andage,salary,job_satisfaction, andwork_life_balance. To compute for the attrition rate, group the dataset by job role. Afterward, you can use thecountfunction to get the frequency of attrition for each job role and then divide it by the total number of observations. Save the computation aspct_attrition. Do not forget to ungroup before storing the output. Store the output asattrition_rate_job_role.Plot for the attrition rate across
job_rolehas been done for you! Study each line of code. You have the freedom to customize your plot accordingly. Show your creativity!
## selecting attrition key variables and save as `attrition_key_var_dta`
attrition_key_var_dta <-
hr_perf_dta |>
select(attrition,
job_role,
department,
age,
salary,
job_satisfaction,
work_life_balance
)
## compute the attrition rate across job_role and save as attrition_rate_job_role
attrition_rate_job_role <-
attrition_key_var_dta |>
group_by(job_role) |>
count(attrition) |>
mutate(pct_attrition = n / sum(n)) |>
ungroup()
## print attrition_rate_job_role
attrition_rate_job_role# A tibble: 24 × 4
job_role attrition n pct_attrition
<chr> <chr> <int> <dbl>
1 Analytics Manager No 185 0.869
2 Analytics Manager Yes 28 0.131
3 Data Scientist No 790 0.570
4 Data Scientist Yes 597 0.430
5 Engineering Manager No 289 0.941
6 Engineering Manager Yes 18 0.0586
7 HR Business Partner No 25 1
8 HR Executive No 90 0.756
9 HR Executive Yes 29 0.244
10 HR Manager No 17 1
# ℹ 14 more rows
## Plot the attrition rate
attrition_rate_job_role |>
mutate(job_role = reorder_within(job_role, pct_attrition, attrition)) |>
ggplot(aes(pct_attrition, job_role, fill = attrition)) +
geom_col(position = "dodge", width = 0.8) +
scale_y_reordered() +
facet_wrap(~ attrition, scales = "free_y", ncol = 1) +
labs(x = "Attrition rate",
y = "Job role")5.2 Identifying attrition key drivers using correlation analysis
Conduct a correlation analysis of key variables:
bi_attrition,salary,years_at_company,job_satisfaction,manager_rating, andwork_life_balance. Use thecor()function to run the correlation analysis. Remove missing values using thena.omit()before running the correlation analysis. Save the output inhr_corr.Use a correlation matrix or heatmap to visualize the relationship between these variables and attrition. You can use the
GGallypackage and use theggcorrfunction to visualize the correlation heatmap. You may explore this site for more information: ggcorr.Discuss which factors seem most correlated with attrition and what that suggests aobut why employees are leaving.
## conduct correlation of key variables.
hr_corr <-
hr_perf_dta |>
select(bi_attrition, salary, years_at_company, job_satisfaction, manager_rating, work_life_balance) |>
na.omit() |>
cor()
## print hr_corr
hr_corr bi_attrition salary years_at_company job_satisfaction
bi_attrition 1.000000000 -0.211181478 -0.6896527798 0.0132368129
salary -0.211181478 1.000000000 0.2206442116 0.0053054850
years_at_company -0.689652780 0.220644212 1.0000000000 0.0008700583
job_satisfaction 0.013236813 0.005305485 0.0008700583 1.0000000000
manager_rating -0.007654429 -0.001596736 0.0178656879 -0.0158205481
work_life_balance 0.003428836 -0.001517145 0.0079339508 0.0417242942
manager_rating work_life_balance
bi_attrition -0.007654429 0.003428836
salary -0.001596736 -0.001517145
years_at_company 0.017865688 0.007933951
job_satisfaction -0.015820548 0.041724294
manager_rating 1.000000000 0.007996938
work_life_balance 0.007996938 1.000000000
## install GGally package and use ggcorr function to visualize the correlation
library(GGally)
ggcorr(hr_corr, label = TRUE)Provide your discussion here.
5.3 Predictive modeling for attrition
Create a logistic regression model to predict employee attrition using the following variables:
salary,years_at_company,job_satisfaction,manager_rating, andwork_life_balance. Save the model ashr_attrition_glm_model. Print the summary of the model using thesummaryfunction.Install the
sjPlotpackage and use thetab_modelfunction to display the summary of the model. You may read the documentation here on how to customize your model summary.Also, use the
plot_modelfunction to visualize the model coefficients. You may read the documentation here on how to customize your model visualization.Discuss the results of the logistic regression model and what they suggest about the factors that contribute to employee attrition.
## run a logistic regression model to predict employee attrition
## save the model as hr_attrition_glm_model
hr_attrition_glm_model <-
glm(bi_attrition ~ salary + years_at_company + job_satisfaction + manager_rating + work_life_balance,
data = hr_perf_dta,
family = "binomial")
## print the summary of the model using the summary function
summary(hr_attrition_glm_model)
Call:
glm(formula = bi_attrition ~ salary + years_at_company + job_satisfaction +
manager_rating + work_life_balance, family = "binomial",
data = hr_perf_dta)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 2.571e+00 2.173e-01 11.831 <2e-16 ***
salary -3.633e-06 4.086e-07 -8.893 <2e-16 ***
years_at_company -6.333e-01 1.476e-02 -42.919 <2e-16 ***
job_satisfaction 3.470e-02 3.186e-02 1.089 0.276
manager_rating 5.071e-03 3.810e-02 0.133 0.894
work_life_balance 2.587e-02 3.198e-02 0.809 0.419
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 8574.5 on 6708 degrees of freedom
Residual deviance: 4781.6 on 6703 degrees of freedom
(190 observations deleted due to missingness)
AIC: 4793.6
Number of Fisher Scoring iterations: 5
## install sjPlot package and use tab_model function to display the summary of the model
library(sjPlot)
tab_model(hr_attrition_glm_model)| bi attrition | |||
| Predictors | Odds Ratios | CI | p |
| (Intercept) | 13.08 | 8.56 – 20.07 | <0.001 |
| salary | 1.00 | 1.00 – 1.00 | <0.001 |
| years at company | 0.53 | 0.52 – 0.55 | <0.001 |
| job satisfaction | 1.04 | 0.97 – 1.10 | 0.276 |
| manager rating | 1.01 | 0.93 – 1.08 | 0.894 |
| work life balance | 1.03 | 0.96 – 1.09 | 0.419 |
| Observations | 6709 | ||
| R2 Tjur | 0.502 | ||
## use plot_model function to visualize the model coefficients
plot_model(hr_attrition_glm_model)Provide your discussion here.
5.4 Analysis of compensation and turnover
Compare the average monthly income of employees who left the company (
bi_attrition = 1) and those who stayed (bi_attrition = 0). Use thet.testfunction to conduct a t-test and determine if there is a significant difference in average monthly income between the two groups. Save the results in a variable calledattrition_ttest_results.Install the
reportpackage and use thereportfunction to generate a report of the t-test results.Install the
ggstatsplotpackage and use theggbetweenstatsfunction to visualize the distribution of monthly income for employees who left and those who stayed. Make sure to map thebi_attritionvariable to thexargument and thesalaryvariable to theyargument.Visualize the
salaryvariable for employees who left and those who stayed usinggeom_histogramwithgeom_freqpoly. Make sure to facet the plot by thebi_attritionvariable and applyalphaon the histogram plot.Provide recommendations on whether revising compensation policies could be an effective retention strategy.
## compare the average monthly income of employees who left and those who stayed
attrition_ttest_results <-
t.test(salary ~ bi_attrition, data = hr_perf_dta)
## print the results of the t-test
attrition_ttest_results
Welch Two Sample t-test
data: salary by bi_attrition
t = 18.869, df = 5524.2, p-value < 2.2e-16
alternative hypothesis: true difference in means between group 0 and group 1 is not equal to 0
95 percent confidence interval:
38577.82 47523.18
sample estimates:
mean in group 0 mean in group 1
125007.26 81956.76
## install the report package and use the report function to generate a report of the t-test results
library(report)
report(attrition_ttest_results)Effect sizes were labelled following Cohen's (1988) recommendations.
The Welch Two Sample t-test testing the difference of salary by bi_attrition
(mean in group 0 = 1.25e+05, mean in group 1 = 81956.76) suggests that the
effect is positive, statistically significant, and medium (difference =
43050.50, 95% CI [38577.82, 47523.18], t(5524.24) = 18.87, p < .001; Cohen's d
= 0.51, 95% CI [0.45, 0.56])
# install ggstatsplot package and use ggbetweenstats function to visualize the distribution of monthly income for employees who left and those who stayed
library(ggstatsplot)
ggbetweenstats(data = hr_perf_dta, x = bi_attrition, y = salary)# create histogram and frequency polygon of salary for employees who left and those who stayed
hr_perf_dta |>
ggplot(aes(x = salary)) +
geom_histogram(alpha = 0.3) +
geom_freqpoly() +
facet_wrap(~ bi_attrition) +
labs(x = "Attrition",
y = "Monthly income")Provide your discussion here.
5.5 Employee satisfaction and performance analysis
Analyze the average performance ratings (both
ManagerRatingandSelfRating) of employees who left vs. those who stayed. Use thegroup_byandcountfunctions to calculate the average performance ratings for each group.Visualize the distribution of
SelfRatingfor employees who left and those who stayed using a bar plot. Use theggplotfunction to create the plot and map theSelfRatingvariable to thexargument and thebi_attritionvariable to thefillargument.Similarly, visualize the distribution of
ManagerRatingfor employees who left and those who stayed using a bar plot. Make sure to map theManagerRatingvariable to thexargument and thebi_attritionvariable to thefillargument.Create a boxplot of
salarybyjob_satisfactionandbi_attritionto analyze the relationship between salary, job satisfaction, and attrition. Use thegeom_boxplotfunction to create the plot and map thesalaryvariable to thexargument, thejob_satisfactionvariable to theyargument, and thebi_attritionvariable to thefillargument. You need to transform thejob_satisfactionandbi_attritionvariables into factors before creating the plot or within theggplotfunction.Discuss the results of the analysis and provide recommendations for HR interventions based on the findings.
# Analyze the average performance ratings (both ManagerRating and SelfRating) of employees who left vs. those who stayed.
hr_perf_dta |>
na.omit() %>%
group_by(bi_attrition) |>
summarise(avg_manager_rating = mean(manager_rating, na.rm = TRUE),
avg_self_rating = mean(self_rating, na.rm = TRUE))# A tibble: 2 × 3
bi_attrition avg_manager_rating avg_self_rating
<dbl> <dbl> <dbl>
1 0 3.48 3.98
2 1 3.46 3.99
# Visualize the distribution of SelfRating for employees who left and those who stayed using a bar plot.
hr_perf_dta |>
na.omit() %>%
ggplot(aes(x = self_rating, fill = factor(bi_attrition))) +
geom_bar(position = "dodge") +
labs(x = "Self Rating",
y = "Count",
fill = "Attrition")# Visualize the distribution of ManagerRating for employees who left and those who stayed using a bar plot.
hr_perf_dta |>
na.omit() %>%
ggplot(aes(x = manager_rating, fill = factor(bi_attrition))) +
geom_bar(position = "dodge") +
labs(x = "Manager Rating",
y = "Count",
fill = "Attrition")# create a boxplot of salary by job_satisfaction and bi_attrition to analyze the relationship between salary, job satisfaction, and attrition.
hr_perf_dta |>
na.omit() %>%
ggplot(aes(y = factor(job_satisfaction), x = salary, fill = factor(bi_attrition))) +
geom_boxplot() +
labs(x = "Job Satisfaction",
y = "Salary",
fill = "Attrition")Provide your discussion here.
5.6 Work-life balance and retention strategies
At this point, you are already well aware of the dataset and the possible factors that contribute to employee attrition. Using your R skills, accomplish the following tasks:
Analyze the distribution of WorkLifeBalance ratings for employees who left versus those who stayed.
Use visualizations to show the differences.
Assess whether employees with poor work-life balance are more likely to leave.
You have the freedom how you will accomplish this task. Be creative and provide insights that will help HR develop effective retention strategies.
5.7 Recommendations for HR interventions
Based on the analysis conducted, provide recommendations for HR interventions that could help reduce employee attrition and improve overall employee satisfaction and performance. You may use the following question as guide for your recommendations and discussions.
What are the key factors contributing to employee attrition in the company?
Which factors are most strongly correlated with attrition?
What strategies could be implemented to improve employee retention and satisfaction?
How can HR leverage the insights from the analysis to develop effective retention strategies?
What are the potential benefits of implementing these strategies for the company?